vllm.cpp is a 1:1 C++ port of vLLM in a 66 MiB binary with token-identical output.

mudler/vllm.cpp mirrors vLLM's V1 / Model Runner V2 architecture in pure C++ with no Python, PyTorch, or ggml at runtime, shipping a shared/static libvllm with a stable 17-symbol C ABI, an example CLI, and an OpenAI-compatible server. Install footprint 66 MiB against vLLM's 9.1 GiB. Cold start 36.5s versus 221.5s, a 6.1x improvement that decides whether scale-to-zero is viable. On Qwen3.6-27B on an NVIDIA GB10 it's token-identical to vLLM at every concurrency measured while slightly faster (86.05 vs 82.32 tok/s at concurrency 1, 1,095 vs 1,076 at 32), and 1.18x faster than llama.cpp on CPU prefill. 25+ architectures supported but most non-Qwen dense models are speed-pending, with documented gaps in vision and speculative decoding. 249 stars.